WEBVTT

00:00:04.890 --> 00:00:07.610
From the New York Times,
I'm Michael Bavaro.

00:00:07.740 --> 00:00:12.820
This is The Daily.
Last night, in a lavish ceremony,

00:00:13.230 --> 00:00:17.130
President Trump welcomed
Chinese leader Xi Jinping

00:00:17.340 --> 00:00:22.750
As he arrived in the United States, for
two days of closely watched negotiations.

00:00:25.990 --> 00:00:30.570
On their crowded agenda,
what if any role they should play

00:00:30.720 --> 00:00:33.500
in regulating artificial intelligence.

00:00:33.690 --> 00:00:37.207
amid a global freakout over its risks.

00:00:37.290 --> 00:00:39.350
Thank you. Thank you. today.

00:00:39.960 --> 00:00:42.600
As my colleague David Sanger explains...

00:00:42.900 --> 00:00:47.740
When it comes to protecting humanity
from technological catastrophe,

00:00:48.000 --> 00:00:53.580
there's really only one proven
model for the two leaders to look to.

00:00:56.240 --> 00:00:59.820
It's Thursday, September 24th.

00:01:03.000 --> 00:01:03.880
David?

00:01:04.500 --> 00:01:05.560
Up from Washington.

00:01:06.510 --> 00:01:09.150
DC sitting here in our New York studios.

00:01:09.750 --> 00:01:12.510
And I didn't even have
to come to you this time.

00:01:13.170 --> 00:01:17.816
I just had to come see the Dailies
Empire in New York. In person? In person.

00:01:17.899 --> 00:01:22.690
Yeah. It's pretty remarkable. It's
a humble empire. It's a late stage empire.

00:01:24.470 --> 00:01:29.947
So, David, when President Trump and the
leader of China, Xi Jinping, meet China,

00:01:30.030 --> 00:01:35.031
In Washington today, they're going
to broach this subject on many people's

00:01:35.115 --> 00:01:40.047
minds, which is the risk of artificial
intelligence, that risk including,

00:01:40.130 --> 00:01:42.757
of course, these swarming rogue people.

00:01:42.840 --> 00:01:48.170
AI agents that have been hacking into
companies and basically breaking the rules

00:01:48.254 --> 00:01:51.840
of their creators.
And what she and Trump uniquely...

00:01:52.050 --> 00:01:55.270
can maybe do about it
is going to be the subject of

00:01:55.410 --> 00:02:00.127
at least some of this meeting. And I
think we've all come to understand that,

00:02:00.210 --> 00:02:02.010
The leaders of the U.S. and China...

00:02:02.100 --> 00:02:06.153
are uniquely positioned to talk about the
safety of AI because they're the countries

00:02:06.237 --> 00:02:07.327
with the most advanced

00:02:07.410 --> 00:02:12.607
frontier AI laboratories. And the
only way that AI gets contained is,

00:02:12.690 --> 00:02:17.944
is if both of those countries somehow
act together and at the same time.

00:02:18.027 --> 00:02:22.370
And as this meeting, approached
and this is why we love you so much

00:02:22.800 --> 00:02:25.340
You reached out to us to say, you know...

00:02:25.620 --> 00:02:32.107
I have seen this movie before. Big fears,
big thoughts about coordination, anxiety.

00:02:32.190 --> 00:02:34.327
and solutions to big problems.

00:02:34.410 --> 00:02:37.030
You told us there's a historical parallel.

00:02:37.470 --> 00:02:41.040
to all of this and maybe
a historical solution.

00:02:41.550 --> 00:02:45.416
Well, Michael, there
is an obvious parallel.

00:02:45.499 --> 00:02:52.230
We had an era starting in the late
1940s and through the Cold War where the

00:02:52.440 --> 00:02:55.859
Two states, the Soviet
Union and the United States,

00:02:55.942 --> 00:03:00.390
had created a potentially terrifying
technology that had all kinds

00:03:00.474 --> 00:03:05.334
of commercial uses but could also
bring about civilizational extinction.

00:03:05.418 --> 00:03:10.828
Nuclear weapons. Nuclear weapons. And
everybody had to go figure out how we were

00:03:10.911 --> 00:03:15.771
going to get the benefits without
getting the terror. And you know what?

00:03:15.855 --> 00:03:17.640
It took decades to happen.

00:03:22.440 --> 00:03:29.129
Came up with agreements that ultimately
have brought us to the remarkable moment

00:03:29.213 --> 00:03:35.394
that there has been no use of a nuclear
weapon in anger since 1945. Right.

00:03:35.477 --> 00:03:40.980
Nuclear nonproliferation worked.
Nuclear nonproliferation worked.

00:03:41.070 --> 00:03:44.670
combined with a balance of terror worked.

00:03:44.850 --> 00:03:50.560
So now, quite naturally, everybody's
looking to that model as we begin to think

00:03:50.643 --> 00:03:54.373
about how you might contain
artificial intelligence,

00:03:54.457 --> 00:03:59.653
particularly after this summer
of surprises that we've had where we have

00:03:59.737 --> 00:04:04.870
rogue agents breaking out of their
sandboxes. Right. Treat AI the way.

00:04:05.250 --> 00:04:06.790
superpowers treated.

00:04:06.960 --> 00:04:12.191
nuclear weaponry. Well, look, if you think
that the heads of some of these frontier

00:04:12.275 --> 00:04:17.058
companies are coming out and saying
there was a 10% chance of civilizational

00:04:17.141 --> 00:04:21.924
extinction, right? That's probably
a higher proportion chance than there was

00:04:22.008 --> 00:04:27.047
in the early days of the nuclear era,
that we would wipe each other off the map,

00:04:27.130 --> 00:04:30.460
at least its equivalent.
And so now the question is,

00:04:30.960 --> 00:04:35.166
Can two very different countries
in very different times,

00:04:35.249 --> 00:04:41.185
China and the United States, come to some
kind of framework agreement that could

00:04:41.268 --> 00:04:44.880
begin to get us there?
Okay. Well, given all of…

00:04:45.120 --> 00:04:45.953
That.

00:04:46.830 --> 00:04:50.010
In your capacity, David,
as a historian of history,

00:04:50.190 --> 00:04:56.380
all the world's great technological
dangers, I think it would be instructive,

00:04:56.670 --> 00:04:58.810
to have you tell us the story

00:04:58.950 --> 00:05:03.363
of nuclear nonproliferation, how that old
framework that kept nuclear weapons

00:05:03.447 --> 00:05:07.568
from going off around the world could,
in theory, apply right now to AI.

00:05:07.651 --> 00:05:10.488
So just give us a quick
history lesson here. Well,

00:05:10.571 --> 00:05:14.985
let's just start with when the U.S.
developed its first atomic weapon through

00:05:15.068 --> 00:05:19.656
the Manhattan Project. Familiar to anybody
who's been off to see the Oppenheimer

00:05:19.740 --> 00:05:23.010
movie or read some of this history.
It was in the 1940s.

00:05:28.950 --> 00:05:33.316
another country would beat us to it.
But it wasn't the Soviets at that time.

00:05:33.400 --> 00:05:36.888
It was Germany. In a last
desperate attempt to escape defeat,

00:05:36.971 --> 00:05:40.693
German science threw everything
it had into the perfection of V2.

00:05:40.777 --> 00:05:43.470
Where the fear of American
intelligence was...

00:05:43.560 --> 00:05:47.282
the Germans would get a nuclear
weapon before the United States.

00:05:47.365 --> 00:05:51.443
And there was this just abiding fear
we had to get there first. Right.

00:05:51.527 --> 00:05:55.248
In the spreading, teeming
laboratories of the Manhattan Project,

00:05:55.332 --> 00:05:56.580
the pace intensified.

00:05:56.670 --> 00:06:01.820
Groves had ordered that the first form
must be ready for testing in mid-July.

00:06:01.903 --> 00:06:03.670
And in the summer of 1945,

00:06:04.140 --> 00:06:07.580
Scientists from Los Alamos
went out to do the Trinity test,

00:06:07.663 --> 00:06:12.680
which was the first explosion of a nuclear
bomb to see if the technology could work.

00:06:22.290 --> 00:06:25.057
would never see again. There were...

00:06:25.140 --> 00:06:29.852
People within the Los Alamos program
who were doing mathematical calculations

00:06:29.936 --> 00:06:34.585
and coming to the conclusion that we
could light the entire atmosphere afire

00:06:34.669 --> 00:06:37.160
and destroy the earth in doing the test.

00:06:39.820 --> 00:06:46.820
But despite those fears,
they went ahead with the test.

00:06:51.100 --> 00:06:52.660
Just weeks later.

00:06:55.140 --> 00:06:59.440
American airplane dropped
one bomb on Hiroshima.

00:06:59.760 --> 00:07:06.760
came Hiroshima. And days after that,

00:07:14.580 --> 00:07:19.596
Nagasaki. Four square miles of the city
were blasted into dust and a black rain

00:07:19.679 --> 00:07:23.100
of destruction hovered over
the ruins for many hours.

00:07:23.220 --> 00:07:27.870
And in total, Those two bombs
killed around 200,000 people.

00:07:28.520 --> 00:07:31.768
Of course, the world had never
seen anything like this before.

00:07:31.851 --> 00:07:34.860
Tens of thousands of homes
and buildings were wiped out.

00:07:35.760 --> 00:07:40.469
the devastation was everywhere.
Japan was driven to quick surrender.

00:07:40.552 --> 00:07:46.260
For the future, the choice is peace or
total destruction. The atomic age is here.

00:07:54.840 --> 00:07:59.830
you But somewhat astonishingly,
it didn't stop the arms race.

00:08:00.210 --> 00:08:04.677
Of course, it turned out to be just
the beginning. And now, of course,

00:08:04.760 --> 00:08:10.007
the U.S. thought it had an insurmountable
lead over countries that didn't yet have

00:08:10.090 --> 00:08:13.580
nuclear weapons.
That lead didn't last very long.

00:08:13.770 --> 00:08:18.985
The Soviets were working away on this.
They had a bunch of atomic spies along

00:08:19.069 --> 00:08:21.944
the way who helped them to it.
And by 1949,

00:08:22.028 --> 00:08:25.950
they had conducted a test
and we were off into the races.

00:08:26.370 --> 00:08:31.736
And the 1950s were a period of great
terror where each side was not only

00:08:31.819 --> 00:08:37.336
building up its arsenal, but building
up the powers of these weapons as if

00:08:37.420 --> 00:08:42.483
the weapons used at Hiroshima
and Nagasaki were not powerful enough.

00:08:42.566 --> 00:08:47.856
And by this stage, proliferation
was starting. Off northwest Australia,

00:08:47.939 --> 00:08:53.010
this was the place chosen to explode
Britain's first atomic weapon.

00:08:56.370 --> 00:09:00.815
Britain and France had nuclear weapons.
China wasn't that far behind.

00:09:00.898 --> 00:09:04.770
I've just been told that China
has exploded an atomic bomb.

00:09:05.070 --> 00:09:07.147
Right, things are starting to spin.

00:09:07.230 --> 00:09:10.230
Kind of out of control.
They are. And then we had...

00:09:10.410 --> 00:09:16.147
one big terrifying incident. The purpose
of these bases can be none other

00:09:16.230 --> 00:09:20.448
than to provide a nuclear strike
capability against the Western Hemisphere.

00:09:20.531 --> 00:09:22.687
And it was the Cuban Missile Crisis...

00:09:22.770 --> 00:09:27.562
in October of 1962. This was when U.S.
intelligence came up with evidence

00:09:27.645 --> 00:09:31.235
that the Soviets were moving
nuclear weapons onto Cuba,

00:09:31.318 --> 00:09:34.307
that they would be just
90 miles from Florida,

00:09:34.390 --> 00:09:38.114
that they could easily reach
any place on the East Coast.

00:09:38.197 --> 00:09:43.256
The waterfront in Havana and along other
parts of the coast bristled with gun

00:09:43.339 --> 00:09:48.866
emplacements as the Cuban regime waited to
see what their bosses in the Kremlin were

00:09:48.949 --> 00:09:49.783
to do.

00:09:52.770 --> 00:09:55.724
This was intolerable.
But the question came,

00:09:55.808 --> 00:09:59.927
is this worth going to war over?
Rockets, bombs, ammunition.

00:10:00.010 --> 00:10:04.230
Air support for any limited war
ground operation which might develop.

00:10:04.313 --> 00:10:07.930
Within 12 hours, all units
were assembled and ready to go.

00:10:08.170 --> 00:10:09.430
and

00:10:09.520 --> 00:10:15.467
There were moments when Kennedy thought,
and many around him thought, that...

00:10:15.550 --> 00:10:20.350
The United States was going to disappear,
you know, in the next night or two. Wow.

00:10:26.170 --> 00:10:29.434
A final chapter in the Cuban
crisis is brought to a close.

00:10:29.517 --> 00:10:32.207
As President Kennedy
holds a news conference...

00:10:32.290 --> 00:10:35.147
But in the end, we know how this ended.

00:10:35.230 --> 00:10:38.470
I have today been informed
by Chairman Khrushchev

00:10:38.920 --> 00:10:43.937
that all of the IL-28 bombers now
in Cuba will be withdrawn in 30 days.

00:10:44.020 --> 00:10:48.520
The nuclear crisis
was averted at the last moment.

00:10:48.610 --> 00:10:52.608
And Kennedy then went on to
give a really famous speech.

00:10:52.691 --> 00:10:57.564
It was a commencement speech
at American University in June of 1963.

00:10:57.647 --> 00:11:01.135
So it was just months
before he was assassinated.

00:11:01.218 --> 00:11:03.550
I realized the pursuit of peace.

00:11:04.150 --> 00:11:07.640
is not as dramatic. as the pursuit of war

00:11:07.840 --> 00:11:13.629
And he basically said, look, the terror
of this event has made us think we cannot

00:11:13.712 --> 00:11:14.800
live with this.

00:11:17.800 --> 00:11:19.520
heirs that we all inhabit,

00:11:19.660 --> 00:11:24.944
this small planet. And he said, our most
basic common link is that we all inhabit

00:11:25.027 --> 00:11:28.540
this small planet.
We all breathe the same air.

00:11:28.660 --> 00:11:31.060
We all cherish our children's futures.

00:11:31.210 --> 00:11:34.130
And we are all mortal.
And we are all mortal.

00:11:34.930 --> 00:11:38.590
And that's the line that
has resonated ever since.

00:11:39.490 --> 00:11:44.909
He announced at that moment that there
would be negotiations for a first tiny

00:11:44.993 --> 00:11:49.810
step. First, Chairman Khrushchev,
Prime Minister McMillan and I have agreed

00:11:50.380 --> 00:11:53.740
that high-level discussions
will shortly begin in Moscow,

00:11:54.310 --> 00:11:59.690
looking towards early agreement
on a comprehensive test ban treaty.

00:11:59.920 --> 00:12:03.077
He called it the Comprehensive
Nuclear Test Ban Treaty.

00:12:03.160 --> 00:12:06.400
Our hopes must be tempered
with the caution of history.

00:12:06.850 --> 00:12:08.830
But with our hopes go the hopes...

00:12:08.920 --> 00:12:10.240
of all mankind.

00:12:10.570 --> 00:12:15.782
That the U.S. at that moment, he said,
would stop testing in the atmosphere above

00:12:15.865 --> 00:12:20.358
ground. You know, those you've seen
the images of the mushroom clouds,

00:12:20.441 --> 00:12:25.196
the mushroom clouds. Right. And that
over time we had to move to reduction

00:12:25.279 --> 00:12:28.090
of nuclear weapons.
But let's start simple.

00:12:28.450 --> 00:12:33.884
with testing, with freezes, and with
the first discussion of how you'd keep

00:12:33.967 --> 00:12:37.709
this technology from spreading
to many other states.

00:12:37.792 --> 00:12:43.373
Because Kennedy himself had said that he
thought there could be 15 or even 20

00:12:43.457 --> 00:12:49.590
nuclear-armed nations coming. by the
mid 70s. Wow. 15 or 20. That's right.

00:12:49.870 --> 00:12:54.589
And you began to see a real cultural
awakening in the United States about

00:12:54.673 --> 00:12:59.129
the depth of this danger. Here's
Tony going to his Cub Scout meeting.

00:12:59.212 --> 00:13:03.537
Tony knows the bomb can explode
any time of the year, day or night.

00:13:03.620 --> 00:13:04.870
He is ready for it.

00:13:04.960 --> 00:13:10.004
So, of course, this is the era when the
government is encouraging schools to do

00:13:10.087 --> 00:13:14.500
those ridiculous duck and cover exercises.
Duck and cover exercises.

00:13:14.740 --> 00:13:18.060
Attaboy, Tony. That flash means act fast.

00:13:18.670 --> 00:13:21.615
Of course, everybody knew
that none of this would help.

00:13:21.698 --> 00:13:24.038
Older people will help
us as they always do,

00:13:24.121 --> 00:13:27.810
but there might not be any grown-ups
around when the bomb explodes.

00:13:28.570 --> 00:13:30.380
Then... You're on your own.

00:13:31.030 --> 00:13:36.815
And, you know, this was when Tom Lehrer,
the famous songwriter and satirist,

00:13:36.899 --> 00:13:41.757
had, you know, his song. So long,
mom. I'm off to drop the bomb.

00:13:41.840 --> 00:13:46.467
So don't wait up for me.
And who's next? France got the bomb,

00:13:46.551 --> 00:13:50.251
but don't you grieve
because they're on our side.

00:13:50.334 --> 00:13:53.957
I believe China got the bomb,
but have no fears.

00:13:54.041 --> 00:13:57.670
They can't wipe us out
for at least five years.

00:14:01.030 --> 00:14:06.914
countries are going to seek their next
weapons. Who's next? Who's next?

00:14:06.997 --> 00:14:08.930
Who's next? Who's next?

00:14:12.670 --> 00:14:14.770
And I think this created...

00:14:14.890 --> 00:14:20.165
and incredible political pressure on the
Johnson administration to get this thing

00:14:20.248 --> 00:14:22.630
done. And in 1968, it came together.

00:14:24.480 --> 00:14:28.351
After endless diplomatic
meetings back and forth,

00:14:28.435 --> 00:14:34.808
occupying lots of days at the Waldorf
Astoria Hotel in New York and other places

00:14:34.891 --> 00:14:41.426
around the world before they came up with
what became the Nuclear Nonproliferation

00:14:41.509 --> 00:14:45.623
Treaty. "Distinguished guests,
ladies and gentlemen,

00:14:45.706 --> 00:14:51.840
this is a very reassuring and hopeful
moment in the relations among nations.

00:14:52.030 --> 00:14:58.277
And on July 1st, 1968,
Five years after that Kennedy speech,

00:14:58.360 --> 00:15:01.641
The treaty was signed by President
Johnson at the White House.

00:15:01.724 --> 00:15:04.980
We have come here today to the
East Room of the White House.

00:15:05.410 --> 00:15:11.493
to sign a treaty which limits
the spread of nuclear weapons.

00:15:11.576 --> 00:15:15.912
And then it was signed
in the Soviet Union,

00:15:15.995 --> 00:15:22.995
and then by about 60 countries in total.
Man can still shape his destiny in

00:15:23.703 --> 00:15:28.430
the nuclear age and learn
to live as brothers.

00:15:30.130 --> 00:15:32.981
And here's what the treaty says.
First of all,

00:15:33.064 --> 00:15:37.573
it says we understand there are five
countries that already have nuclear

00:15:37.656 --> 00:15:42.038
weapons and they all commit
to eliminating their stockpiles over time.

00:15:42.121 --> 00:15:44.290
It doesn't specify when. And then.

00:15:44.440 --> 00:15:48.527
Countries that do not have nuclear weapons
will commit that they will never die.

00:15:48.610 --> 00:15:52.210
build them. But in return, they get

00:15:52.540 --> 00:15:57.561
help for peaceful nuclear technology to
build nuclear power plants and so forth.

00:15:57.644 --> 00:16:02.155
And then, of course, there needs
to be an inspection regime to make sure

00:16:02.238 --> 00:16:07.259
that they're not cheating. And that was
the creation of the International Atomic

00:16:07.342 --> 00:16:11.151
Energy Agency, which is based
in Vienna. It's part of the UN.

00:16:11.234 --> 00:16:16.255
And they begin the process of developing
teams that go visit countries that have

00:16:16.338 --> 00:16:17.172
signed this

00:16:22.540 --> 00:16:24.660
being diverted off to bomb projects.

00:16:26.100 --> 00:16:31.578
And As we look back at this, this was even
a bigger moment than we knew at the time.

00:16:31.662 --> 00:16:33.810
What do you mean? Because...

00:16:34.000 --> 00:16:39.749
Nations that seem to be inevitably
racing to adopt a fairly non-complicated

00:16:39.832 --> 00:16:44.571
technology to build nuclear
weapons all agreed to halt. Right.

00:16:44.654 --> 00:16:47.920
There's very few moments in history where.

00:16:48.220 --> 00:16:49.053
Everybody.

00:16:49.840 --> 00:16:54.377
who wants to catch up with the leaders
in an arms race, voluntarily...

00:16:54.460 --> 00:16:57.848
abandon the pursuit.
That's right. Almost everybody.

00:16:57.931 --> 00:17:01.520
There are a couple of outliers
and Israel never signed.

00:17:01.603 --> 00:17:04.657
And while it's never
admitted to having a bomb,

00:17:04.741 --> 00:17:08.796
it's not exactly the biggest
secret in the Middle East. Right.

00:17:08.880 --> 00:17:12.334
North Korea signed but renounced
it in the mid 1990s.

00:17:12.418 --> 00:17:15.672
And they now have an arsenal
of 60 to 100 weapons,

00:17:15.756 --> 00:17:18.960
depending on who you believe.
India and Pakistan

00:17:24.460 --> 00:17:28.300
worry about Iran, but they don't
actually have a nuclear weapon yet.

00:17:28.383 --> 00:17:30.460
But ultimately, this treaty prevents

00:17:30.580 --> 00:17:37.166
the kind of spreading of nuclear weaponry
that Kennedy so worried about. It did.

00:17:37.249 --> 00:17:40.500
And it's remarkable. Kennedy was wrong.

00:17:40.630 --> 00:17:47.270
We've only got nine nuclear states, and
many of these countries have had nuclear

00:17:47.353 --> 00:17:53.152
weapons for decades. And so far,
it has remained relatively contained,

00:17:53.236 --> 00:17:59.707
which is why this does seem like perhaps
the best model we have to think about

00:17:59.790 --> 00:18:05.086
a comparable international
regulatory system for restraining AI.

00:18:05.169 --> 00:18:07.690
That's exactly right, Michael.

00:18:07.900 --> 00:18:14.900
And as everybody enters this early stage
debate about whether or not you can do

00:18:17.270 --> 00:18:24.270
the same with artificial intelligence,
here is an example, a rare example,

00:18:26.047 --> 00:18:33.047
in which nations have voluntarily
agreed to slow a technology,

00:18:37.900 --> 00:18:41.625
the leaders of the frontier companies.
They're not saying stop it.

00:18:41.709 --> 00:18:45.376
They're not saying let's go out
of business. They're saying, hey,

00:18:45.460 --> 00:18:46.960
let's just slow this down.

00:18:47.110 --> 00:18:52.326
so that we can understand how to control
it before we find ourselves in another

00:18:52.409 --> 00:18:57.491
crisis. Right. In some ways, a more
modest request than what nonproliferation

00:18:57.574 --> 00:19:01.649
was about. You're right, Michael.
It is a more modest request,

00:19:01.733 --> 00:19:04.550
but it's a far more complicated situation.

00:19:10.510 --> 00:19:11.890
We'll be right back.

00:19:17.770 --> 00:19:22.367
So David, understanding as we now
do, that nuclear nonproliferation

00:19:22.450 --> 00:19:27.010
as a model is our best and so
far really our only goal.

00:19:27.160 --> 00:19:31.277
model for AI. In what ways...

00:19:31.360 --> 00:19:35.477
is it quite challenging to apply
that model to AI right now?

00:19:35.560 --> 00:19:38.977
You just said it's very
different and complicated.

00:19:39.060 --> 00:19:43.947
So there was something of a remarkable
simplicity to the arms race with

00:19:44.030 --> 00:19:47.237
the Soviet Union,
even while it was terrifying.

00:19:47.320 --> 00:19:52.640
What was simple about it? Well, two
countries had almost all the technology,

00:19:52.990 --> 00:19:57.495
We were only worried about states. We
were not worried about non-state actors,

00:19:57.578 --> 00:20:01.640
terrorist groups, people
working in their basements and

00:20:01.970 --> 00:20:06.094
If you didn't have plutonium or uranium,
you couldn't make a bomb.

00:20:06.178 --> 00:20:10.130
So we knew that there was a way
to sort of cut off the oxygen.

00:20:10.520 --> 00:20:16.482
There was a limited number of people
with launch authority in each country,

00:20:16.565 --> 00:20:23.011
and both countries had their own political
dynamic for wanting to make sure there

00:20:23.095 --> 00:20:29.140
was no accident and there was no
escalation if they could avoid it. Mm-hmm.

00:20:29.330 --> 00:20:32.982
That didn't mean they were
willing to give up the weapons,

00:20:33.065 --> 00:20:36.327
but it did mean that you
could contain the problem.

00:20:36.410 --> 00:20:41.184
But when you think about AI,
the situation is completely different.

00:20:41.268 --> 00:20:45.110
The governments don't really
control this technology.

00:20:46.130 --> 00:20:49.408
You don't need the equivalent
of uranium and plutonium.

00:20:49.491 --> 00:20:53.441
You just need compute a lot of it.
And maybe that's the oxygen for

00:20:53.524 --> 00:20:58.230
this particular case. Computer power.
You need computer power and tons of it.

00:20:58.370 --> 00:21:04.506
There's no reason to believe that as
this technology becomes more efficient,

00:21:04.590 --> 00:21:10.399
that smaller and smaller groups
working out of basements, terror groups,

00:21:10.482 --> 00:21:16.864
countries that have interest in causing
disruption can't get a hold of basic AI

00:21:16.947 --> 00:21:22.430
technology. And we've already seen
some examples of that happening.

00:21:22.790 --> 00:21:26.561
There's one other big difference,
Michael. In the Cold War,

00:21:26.644 --> 00:21:31.395
we didn't have to convince our national
leaders that this was a terrifying

00:21:31.479 --> 00:21:36.770
technology. They had to, if anything, to
convince their populations of that. But.

00:21:36.860 --> 00:21:42.803
In this case, we have an American
president who is declaring that all these

00:21:42.886 --> 00:21:46.180
warnings are a hoax. And Xi Jinping is...

00:21:46.580 --> 00:21:50.745
is clearly more concerned,
as some of his aides have made clear,

00:21:50.828 --> 00:21:53.997
about the threat to the
Communist Party of China,

00:21:54.081 --> 00:21:56.918
rather than its threat
to the broader world.

00:21:57.001 --> 00:22:00.320
So you're saying the leaders
of the superpowers...

00:22:00.470 --> 00:22:03.218
locked in this arms race,
the AI arms race,

00:22:03.301 --> 00:22:07.959
can't even agree on what the threat is.
Whereas back in the nuclear era,

00:22:08.042 --> 00:22:13.029
the threat was pretty self-evident
and the leaders could agree on it. It was,

00:22:13.112 --> 00:22:16.470
this is going to kill all
of us. That's right. Now,

00:22:16.910 --> 00:22:22.427
They also recognized that they were not
going to give up all nuclear weapons back

00:22:22.511 --> 00:22:26.991
then because they believed they
needed that for national survival.

00:22:27.074 --> 00:22:31.624
In this case, the U.S. and China
both think they need to be pushing

00:22:31.707 --> 00:22:35.081
this technology ahead
for their economic survival.

00:22:35.164 --> 00:22:38.884
But it's also a big part
of the struggle for dominance.

00:22:38.967 --> 00:22:42.747
China's made no secret
of the fact that by 2049, China,

00:22:42.830 --> 00:22:47.050
the 100th anniversary of the creation
of the People's Republic of China.

00:22:47.133 --> 00:22:51.533
It wants to be the biggest military power,
the biggest technological power,

00:22:51.616 --> 00:22:53.230
the biggest economic power.

00:22:54.530 --> 00:23:00.297
And you don't get to that point without
AI. Right. And so if you're China...

00:23:00.380 --> 00:23:06.233
and you're suddenly faced with the
prospect of a nuclear non-proliferation

00:23:06.316 --> 00:23:10.564
style agreement with your
biggest adversary, the U.S.,

00:23:10.648 --> 00:23:15.300
when it comes to AI, you're
probably thinking to yourself,

00:23:16.490 --> 00:23:20.150
If we sign that, then we might be
signing away our economic future.

00:23:21.680 --> 00:23:23.640
Adding to that suspicion...

00:23:24.140 --> 00:23:28.758
would probably be the reality,
and we've covered it a lot on the show,

00:23:28.841 --> 00:23:33.140
that China already does a much
more robust job of regulating AI,

00:23:33.770 --> 00:23:38.052
than the United States does. The U.S.
has taken a really hands-off approach,

00:23:38.135 --> 00:23:41.153
and AI here is even more
advanced than it is in China.

00:23:41.236 --> 00:23:44.510
And so if you're Xi Jinping,
you're thinking to yourself,

00:23:44.750 --> 00:23:46.550
Why on earth would I sign a deal?

00:23:47.090 --> 00:23:48.990
that locks in U.S.

00:23:49.610 --> 00:23:55.539
AI advantages, It makes it harder for us
to catch up and realize our AI future.

00:23:55.623 --> 00:23:57.710
It's exactly right, because...

00:23:58.040 --> 00:24:03.507
The Chinese are convinced the U.S.
is already trying to do that with export

00:24:03.590 --> 00:24:06.920
controls on the world's
fastest chips. Right.

00:24:07.400 --> 00:24:13.765
President Trump has agreed to give the
Chinese the world's second fastest chips,

00:24:13.848 --> 00:24:20.213
but not the fastest. And the Chinese view
of this is this is modern containment.

00:24:20.296 --> 00:24:23.517
This is their effort to preserve a lead.

00:24:23.600 --> 00:24:27.491
And then if you're other countries,
you're probably thinking,

00:24:27.574 --> 00:24:32.247
I don't want to sign anything that's
going to cut off my economic future.

00:24:32.331 --> 00:24:37.069
And if everybody in the world is running
around saying the economic future

00:24:37.152 --> 00:24:40.261
is completely dependent
on the development of AI,

00:24:40.344 --> 00:24:45.147
why would I sign a treaty that would
keep me from being a full participant?

00:24:45.231 --> 00:24:49.137
So we're clearly running up
against some of the real issues.

00:24:49.220 --> 00:24:50.053
limits.

00:24:50.390 --> 00:24:57.388
of the nuclear model for AI. So let me
return to the slightly more optimistic

00:24:57.471 --> 00:25:01.150
thought that it is our only model. So...

00:25:01.250 --> 00:25:02.410
Does that mean...

00:25:02.540 --> 00:25:06.447
that President Xi and President
Trump are in any way inclined to,

00:25:06.530 --> 00:25:11.805
to pursue it? Well, in some ways,
they already have. What do you mean? Well,

00:25:11.889 --> 00:25:17.164
back during the Biden administration,
Biden's advisers wanted to see if they

00:25:17.247 --> 00:25:22.452
could get the Chinese engaged in the
most basic conversation on this topic.

00:25:22.535 --> 00:25:28.445
And it took about a year and a half to get
the Chinese to agree to basically a single

00:25:28.529 --> 00:25:31.490
sentence, which is which is basically that

00:25:31.820 --> 00:25:37.928
Neither country would let artificial
intelligence agents control their nuclear

00:25:38.011 --> 00:25:39.440
weapons stockpile.

00:25:39.800 --> 00:25:44.805
A seemingly very wise pledge. You wouldn't
think it would take a year and a half

00:25:44.888 --> 00:25:49.287
to get to something that basic.
But you're saying the U.S. and China –

00:25:49.370 --> 00:25:51.507
have come to an understanding...

00:25:51.590 --> 00:25:55.110
that feels like an extension of

00:25:55.370 --> 00:25:59.798
the nuclear containment policies that
the U S and Russia and everybody else have

00:25:59.882 --> 00:26:04.760
agreed to. We're going to keep,
AI and nuclear weapons from intermingling.

00:26:05.330 --> 00:26:10.532
That's the concept, Michael. Now, no
one really knows what this sentence means

00:26:10.616 --> 00:26:11.307
because

00:26:11.390 --> 00:26:16.315
Supposing it simply means you're not going
to let AI agents decide whether or not

00:26:16.398 --> 00:26:19.778
to launch nuclear weapons.
OK, we can all agree on that.

00:26:19.861 --> 00:26:24.168
But how do you know whether the
information that President Trump or one

00:26:24.251 --> 00:26:28.372
of his successors is receiving
or President Xi and one of one of his

00:26:28.455 --> 00:26:33.566
successors receives about the threat that
might lead to a decision to employ nuclear

00:26:33.649 --> 00:26:37.050
weapons? It wasn't influenced by AI.
Yeah, faked by AI.

00:26:37.220 --> 00:26:39.100
influenced by AI,

00:26:39.290 --> 00:26:44.066
the result of a quick AI search
by an aide who's sitting outside the room.

00:26:44.150 --> 00:26:48.072
So we don't really know what
this means, but it was progress.

00:26:48.155 --> 00:26:53.063
And you could imagine a situation
in which President Trump and President Xi,

00:26:53.146 --> 00:26:57.200
if they were willing, could build
on this and not just nuclear,

00:26:57.284 --> 00:27:02.060
that we don't want AI anywhere near
the development of biological weapons.

00:27:02.143 --> 00:27:06.132
Right, because AI has now reached
the level of sophistication,

00:27:06.215 --> 00:27:09.117
we have come to understand,
where it could...

00:27:09.200 --> 00:27:16.200
help actually design biological weapons.
And we don't want AI to be used

00:27:16.990 --> 00:27:18.180
in hacking.

00:27:18.350 --> 00:27:25.008
Because we have come to understand that
AI can find and exploit vulnerabilities

00:27:25.091 --> 00:27:29.360
in the electric power grid
or in the water systems,

00:27:29.443 --> 00:27:33.687
that it might take human
hackers months or months.

00:27:33.770 --> 00:27:38.291
or years to find and then
design an attack. In other words,

00:27:38.374 --> 00:27:43.988
that it speeds up cyber attacks
and is pushing us into a new era of cyber

00:27:44.072 --> 00:27:44.930
insecurity.

00:27:45.680 --> 00:27:50.340
So, David, let's just talk logistics and
brass tacks. Is there a world in which...

00:27:50.510 --> 00:27:53.277
the leader of the United
States and the leader of China.

00:27:53.360 --> 00:27:58.811
sit down with their teams and begin
to hash out the kind of nuclear

00:27:58.895 --> 00:28:02.860
nonproliferation scheme
that we're really doing.

00:28:02.960 --> 00:28:09.297
talking about here for the world of AI,
in which they begin to pledge...

00:28:09.380 --> 00:28:14.386
Things like, "We both agree," that AI
shouldn't touch biological weapons,

00:28:14.469 --> 00:28:18.609
shouldn't be involved in the hacking
of private corporations or governments.

00:28:18.693 --> 00:28:24.440
And this all eventually finds its way into
something that resembles a treaty that

00:28:24.980 --> 00:28:29.160
And brings us. To a point. where

00:28:29.900 --> 00:28:32.997
What looks like the nuclear
non-perforation deal exists in AI.

00:28:33.080 --> 00:28:37.621
Well, this week, as President
Xi and President Trump meet, no,

00:28:37.705 --> 00:28:41.425
I don't think there's even
a remote chance of that.

00:28:41.509 --> 00:28:47.467
I think we would be lucky to get a second
sentence with a joint experts committee

00:28:47.551 --> 00:28:51.048
that will meet sometime.
But over the long term,

00:28:51.131 --> 00:28:54.027
I could imagine a broader conversation.

00:28:54.110 --> 00:29:00.435
understanding, not really a treaty,
in which you could set up some mechanisms

00:29:00.518 --> 00:29:01.850
that might work.

00:29:02.540 --> 00:29:03.373
Early warning.

00:29:03.740 --> 00:29:09.405
sharing of models back and forth for a few
days or weeks before they are publicly

00:29:09.488 --> 00:29:14.514
released. Although I think American
companies would be reluctant to give

00:29:14.598 --> 00:29:17.720
the Chinese an early
peak and certainly not.

00:29:17.840 --> 00:29:22.030
The inspection part of the nuclear
nonproliferation treaty,

00:29:22.113 --> 00:29:27.824
you could imagine, you already saw a hint
of that in an essay from Dario Amadei,

00:29:27.908 --> 00:29:31.518
the head of Anthropic,
about a week and a half ago,

00:29:31.602 --> 00:29:36.371
where he described a commitment
by Anthropic to allow in some group

00:29:36.454 --> 00:29:42.238
of independent inspectors to sit among
the developers and watch over the creation

00:29:42.321 --> 00:29:46.160
of new and more powerful
models and blow the whistle.

00:29:47.840 --> 00:29:53.489
that they believed could pose a threat
to humanity. And inside OpenAI,

00:29:53.572 --> 00:29:59.917
they called this the IAEA model, the
International Atomic Energy Agency model.

00:30:00.000 --> 00:30:04.336
That there are these inspectors who are
independent enough that they can call up

00:30:04.419 --> 00:30:08.699
people like you and me and say, hey,
there's a big problem in Anthropic and let

00:30:08.782 --> 00:30:11.019
me come over and explain it to you. Right.

00:30:11.102 --> 00:30:14.554
They're referring literally
to the inspectors who exist to check

00:30:14.637 --> 00:30:18.310
on countries' nuclear programs.
I mean, we keep coming back to this,

00:30:18.393 --> 00:30:21.100
but everywhere you turn,
everybody ultimately is.

00:30:21.510 --> 00:30:27.201
returns to the nuclear nonproliferation
model. That's true. Now,

00:30:27.284 --> 00:30:33.847
do I think this can be done by treaty?
No. Treaties take forever forever.

00:30:33.930 --> 00:30:38.314
to negotiate. And this technology
is racing like a freight train.

00:30:38.397 --> 00:30:43.400
The fundamental nuclear technology
hasn't changed all that much from 1945.

00:30:43.483 --> 00:30:47.950
So another big difference.
here is just simply speed. That's right.

00:30:48.270 --> 00:30:50.640
Hmm. Well... when I'm,

00:30:50.730 --> 00:30:54.674
Sitting here listening to you, I'm also
doing a little bit of a mental calculation

00:30:54.757 --> 00:30:57.960
of the time it took
for the Non-Proliferation Treaty

00:30:58.230 --> 00:31:04.264
to come into existence after the first
nuclear weapon was dropped on Japan.

00:31:04.347 --> 00:31:10.440
And I think it was basically 20
something years. And, That's a long time.

00:31:10.530 --> 00:31:14.230
But it does suggest
that administrations change,

00:31:14.340 --> 00:31:19.480
and viewpoints change, and these
things are painstaking and they're slow,

00:31:19.564 --> 00:31:24.030
but ultimately... Everyone
did agree to sign that treaty.

00:31:24.240 --> 00:31:27.930
So is it foolhardy to think that over time

00:31:28.590 --> 00:31:31.828
people will come around to the
same kind of conclusion with AI,

00:31:31.912 --> 00:31:36.130
and that we could be looking back in 20
years, you'll be really gray, I'll be...

00:31:36.450 --> 00:31:39.561
just a little less gray.
And we'll say, wow, can you believe it?

00:31:39.644 --> 00:31:41.994
They took their time,
but they got there on AI.

00:31:42.720 --> 00:31:46.310
Your question ignores
the issue of time. Look,

00:31:46.393 --> 00:31:51.660
we don't exactly understand
the threats we're dealing with with AI,

00:31:51.743 --> 00:31:56.771
but there are some engineers who
are saying we have just months,

00:31:56.854 --> 00:32:01.802
if a year away from recursive
self-improvement, that is to say,

00:32:01.885 --> 00:32:04.520
code that keeps improving itself.

00:32:04.710 --> 00:32:09.463
or artificial general intelligence
within months or a year or two,

00:32:09.547 --> 00:32:14.750
we don't have 25 years this time.
We may not have two and a half years.

00:32:15.150 --> 00:32:18.827
This time, which then
raises the next question.

00:32:18.910 --> 00:32:25.547
Is it going to take a Cuba missile crisis
like event in AI to rally the world around

00:32:25.630 --> 00:32:29.710
this? And thank you.
I think we'd rather miss that.

00:32:30.600 --> 00:32:36.287
And remember, in the atomic age,
we started with Hiroshima and Nagasaki.

00:32:36.371 --> 00:32:41.500
So there was no debate about
the power of these weapons to kill.

00:32:41.910 --> 00:32:48.729
But this takes us to the biggest
difference between AI and nuclear.

00:32:48.813 --> 00:32:50.770
In the nuclear era,

00:32:51.330 --> 00:32:56.330
We never really doubted that we
all controlled the arsenal.

00:32:56.970 --> 00:33:02.400
You know, at one point,
Winston Churchill in 1955 – said,

00:33:02.790 --> 00:33:09.540
The atomic bomb, with all its terrors, did
not carry us outside the scope of human

00:33:09.623 --> 00:33:10.457
control.

00:33:10.980 --> 00:33:12.260
By which he meant...

00:33:12.870 --> 00:33:18.250
Human beings were making the decisions
and containing the size of the –

00:33:18.420 --> 00:33:19.880
weapon and the reaction.

00:33:20.670 --> 00:33:22.070
In AI,

00:33:23.070 --> 00:33:29.631
We have no assurance of human control.
That once we've discovered that agents

00:33:29.714 --> 00:33:35.585
were hacking into Hugging Face
and into other corporate sites without

00:33:35.669 --> 00:33:38.377
the knowledge of their creators,

00:33:38.460 --> 00:33:44.869
we have begun to question the sort
of Frankenstein's monster nature of

00:33:44.952 --> 00:33:46.900
this whole thing. And

00:33:47.130 --> 00:33:51.450
If that's the case, then we have
ignored the Churchill warning.

00:33:52.170 --> 00:33:57.301
that we are now on the verge
of being outside of human control.

00:33:57.384 --> 00:34:02.350
And that wasn't really much
of an issue during the pandemic.

00:34:02.580 --> 00:34:08.224
nuclear era, but it may be the most
urgent problem in the age of artificial

00:34:08.307 --> 00:34:09.300
intelligence.

00:34:11.480 --> 00:34:14.702
Well, David, thank you,
I guess. Well, Michael,

00:34:14.786 --> 00:34:19.920
we live in fascinating times. Yeah.
Appreciate your time. Thanks. Thanks.

00:34:20.250 --> 00:34:22.310
you

00:34:23.610 --> 00:34:25.670
you

00:34:26.280 --> 00:34:30.240
*music* We'll be right back.

00:34:30.930 --> 00:34:32.560
you you

00:34:38.310 --> 00:34:42.690
Here's what else you need to know today.
On Thursday morning,

00:34:42.773 --> 00:34:47.300
a federal judge found that President
Trump's order barring CNN,

00:34:47.383 --> 00:34:53.080
MSNOW and Politico from the White House
was likely unconstitutional and ordered

00:34:53.163 --> 00:34:56.090
that the president restore their access.

00:34:56.460 --> 00:35:02.137
The judge, a Trump appointee, rejected the
president's claim that revoking the media

00:35:02.220 --> 00:35:07.020
outlet's press credentials was necessary
to protect national security.

00:35:07.860 --> 00:35:11.770
Trump is expected
to appeal the ruling. And:

00:35:11.853 --> 00:35:18.666
Our nation has developed the power
of the government and the consequences of

00:35:18.750 --> 00:35:23.381
the sanctions. In a defiant wartime
speech at the United Nations,

00:35:23.465 --> 00:35:28.518
Iran's president said that his country
would not be bullied into submission

00:35:28.602 --> 00:35:30.040
by the United States.

00:35:34.360 --> 00:35:40.379
President Masood Pazeshkin vowed that
Iran would defend itself until, quote,

00:35:40.463 --> 00:35:45.120
its last breath and would not
give up its nuclear program.

00:35:51.280 --> 00:35:55.954
Today's episode was produced
by Shannon Lynn and Jessica Chung,

00:35:56.037 --> 00:36:01.013
with help from Michael Simon Johnson.
It was edited by Maria Byrne,

00:36:01.097 --> 00:36:05.640
fact-checked by Susan Lee.
Contains music by Will Reed.

00:36:05.730 --> 00:36:10.647
Dan Powell and Pat McCusker,
and was engineered by Chris Wood.

00:36:10.730 --> 00:36:13.230
Our theme music is by Wonderly.

00:36:13.470 --> 00:36:15.970
Special thanks to Daniel Ramirez.

00:36:22.890 --> 00:36:26.730
That's it for The Daily.
I'm Michael Barbaro.

00:36:26.820 --> 00:36:28.080
See you tomorrow.
